A divide and conquer approach for in-depth analysis of brain connectivity network using ordinal sequence based characterizer

• Published in: Multimedia tools and applications Vol. 84; no. 13; pp. 11625 - 11651
• Main Authors: Kose, Mangesh Ramaji, Ahirwal, Mitul Kumar, Atulkar, Mithilesh
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.04.2025; Springer Nature B.V
• Subjects: Algorithms; Brain; Classification; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Electroencephalography; Feature extraction; Functional analysis; Graph theory; Multimedia Information Systems; Schizophrenia; Special Purpose and Application-Based Systems; Structural analysis; Track 3: Biometrics and HCI; WESOES; Weighted brain connectivity network; Average weighted centrality; Classification
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-024-19401-7

The human brain is composed of discrete functional regions that interact with each other to generate cognitive and/or physical activities. The functional analysis of the brain can be effectively performed by characterizing the communication among different brain regions. The earlier approaches for characterizing brain connectivity were unable to consider the strength of connection and the ordinal relation among these connections. The proposed study overcomes the limitations of existing studies and provides a novel divide-and-conquer approach for detailed analysis of brain connectivity networks (BCN). The proposed technique is composed of two phases: i) Divide and ii) Conquer. In the divide phase, each BCN is divided into multiple subgraphs, and each subgraph is then evaluated independently using an advanced network characterizer. A weighted electroencephalogram (EEG) subgraph ordinal edge sequence (WESOES) based characterizer is developed to utilize weight information of edges connecting different brain regions and the ordered relationship between weighted edges. The WESOES-based features are extracted from each subgraph corresponding to the BCN. In the conquer phase, the features extracted from each subgraph are integrated in various combinations. The integrated features are classified using various classification algorithms. The proposed technique is tested on the database containing EEG signals from schizophrenia-diseased and healthy subjects. The proposed study achieves 86% classification accuracy. The structural analysis shows the active connection at the left-frontal, central and temporal brain regions for the diseased subject, whereas for the healthy subjects in the central left-temporal and parietal brain regions.